QlikView Memory Management

    In this webinar we will discuss:

    • How QlikView uses memory
    • How to identify your memory hogs
    • Best practices for efficient memory usage

     

    For best viewing, we recommend you set the highest video resolution

     

    Q&A - QlikView Memory Management

    Q: If you clean the cash memory, the last state of the session keeps the same?

    A: Yes, the Session State is written to the .Shared file and will thus be retrived when the user comes back to QlikView.


    Q: How can you set a specific time to automatically clear the cache using the config file? let say you want to trigger 2 times in a day.

    A: We are not able to set specific times, only a number of times per day, starting at midnight. Setting a value of ClearCacheTimesPerDay=2 will clear the cache at midnight and at midday.


    Q: If I have a dashboard that has a result set from a user selection of 100 customers and I reload that document and the answer to the same user selection is now 120 customers. Will the result set automatically be updated to 120 customers without user action

    A: The next time the election is made, a new calculation wil be made and added to the cache.


    Q: Any crucial changes beetween QV11 and QV12 (new engine) on memory usage?

    A: No crucial difference between QlikView 11.20 and QlikView 12 in this regard.

     

    Q: Qliksense uses the same "Memory Management" or are nay diiferences?

    A: Qlik Sense operates similarly.

     

    Q: If I'm using QV on a NUMA enabled machine, what would you recommend for troubleshooting? Anything in particular?

    A: This all depends on if you are suffering from any problems. NUMA is a setting in BIOS that mmanages memory addressing, and we prefer it to be disbaled at BIOS level to better be able to use our own memory management algorithms. Since not all server manifactures allow for this to be manually controlled, we have introduced a feature in the QVS that makes it NUMA aware (introduced in QlikView 11.2 SR2), enabling it to use a soft switch to emulate NUMA being disabled when it is in fact enabled at BIOS level. This comes at a performace cost however, so if possible, NUMA should be disabled in BIOS.

     

    Q: Is it possible to know the specific part of memory allocated byApplication, by session state, and by cached result set?

    A: Not available in te application no. By observing the memory consumption by the QVS when the document is loaded and then used, some deductions can be made. Adding a granular view of this would be a nice feature request.

     

    Q: When a document is reloaded does this take it out of memory including all associated objects to that document

    A: A reload of a document only replaces the existing version of the document in memory with a new version of the document, the result sets for that document will be kept in cache.

     

    Q: If you have a distributed architecture are the working limits applied to the publisher server or the qlikview server server? What would the recommended working set limit be on a server with 244GB or RAM?

    A: The Working Set limits only applies to the QVS process, they do not limit the reload engines. A distributed setup (an environment where the QlikView Services are distributed over more than one server) means that you can set the Working Set limits higher since there are no competing servioces that wants acces to the same physical memory. For that size of server, I would start with a setting of 80% low, 90% high and test for stability.

     

    Q: Are you suggesting not to change the working set limits if RAM less than 128GB?

    A: Any server size can benefit from adjusting the Working Set limits, as a higher value means more cached result sets nad thus a better end user experience. But as a practical rule, NOT adjusting the Working Set limts on servers bigger than 128 GB of RAM normally leads to a waste of resources.

     

    Q: Why QlikView is very slow on a virtual server? What is the optimal setup and specs for a virtual server?

    A: By definition, QlikView is not slow on Virtual Servers. QlikView is however dependant on direct access to physical resources, meaning that if a Virtual Server is configured to share resources with other guests on the same host, we will be competing against other applications on other guests for the same resorce pool. We recommend using dedicated resource pools in virtual environments to guarantee that QlikView can work with direct access to CPU and RAM.

     

    Q: Hi, We have had 1 case using pivot table, it showed "out of object memory" error message. What was really happening, because you said that calculations or aggregatins is processed in cpu. So is this cpu problem or memory problem? Thank you

    A:  There is no contradiction here. A calculation or aggregation is done by using CPU, and then stored in memory. If there are object calculation limits, they would limit how much memory that calculation can take.

     

    Q: Is it possible to unload only a specific document from RAM?

    A: No. The same Document Timeout applies to all documents on the server.

     

    Q: Hi, how big data you have been working with qlikview? How many records?. Is there any workaround for a very big number of data? for example, 65 mio rows per day. Thank you

    A: We know of QlikView documents containing abowe 2 billion rows of data. The bigger the amopunt, the more important data model becomes. A well designed star schema will go a very long way for large data volumes.


    Q: Is the QlikView Server Cache dropped when a developer starts a reload of a big document in QlikView Desktop ON the server or will the reload crash (if it reaches the leftover RAM limit)?

    A: QlikView Server and QlikView Desktop dp not share the same set of memory. Executing a heavy reload in QlikView Desktop on the same machine that hosts the QlikView Server will possybly lead to a mamory starvation as the two processes will compete for the same memory


    Q: What happends to the cache for a specific application, if that application is reloaded frecuently? Does all the cached results gets invalid when the document is reloaded? If ot, how much can be "reused"?

    A: Before we reach the Working Set limits, all cached result sets will be kept in memory.